1. General

Question 1

Types of structured data

Answer 1

• Usual dataset: Time independent.
• Time series: Ordered over uniform time intervals
• Cross section: Multiple subjects
• Multilevel: Several levels + clusters
• Panel (longitudinal): Multilevel with 2 Dimensions - Cross-sectional over time

Question 2

Characteristics of unstructured data

Answer 2

• Non-numeric (e.g. text data).
• Multi-faceted (multiple information points, e.g. speech recording).
• Concurrent representation (study of different research goals from one dataset, e.g. an email).

Question 3

Types of variables

Answer 3

• Discrete random (finite or countably infinite range)
• Continuous random (range is an interval)
• Nominal / Categorical (unordered names)
• Binary (unordered nominal variable of two categories)
• Ordinal (ranked)
• Numerical (quantitative, scaled either by interval or ratio)

Question 4

Types of ML techniques

Answer 4

• Unsupervised
  Observe data and construct low complexity description e.g. Clustering, dimensional reduction techniques like PCA, discretization…).
• Supervised
  Observe training samples and learn function mapping inputs -> outputs
  e.g. regression, classification, ranking…
• Reinforcement Learning
  Learning a map system
  states -> actions, under maximization of cumulative reward

Question 5

Types of missing values in data cleaning

Answer 5

• Truly Missing (no value)
• Coded (value, with a different meaning)
• Miscoded (value, with wrong meaning)

Missing values can be …
MCAR: completely random and unpredictable
MAR: predictable
Missing, dependent on unobserved variables

—> Imputation (substitution) by Mean, Median, Stratified (sorting), Regression

Question 6

Big Data Characteristics

Answer 6

• Explosion of data volume and velocity from different domains
• Cheaper storage availability
• Faster and cheaper computation power
• Analysis of unstructured data

Question 7

ML vs. Statistics

Answer 7

Statistics vs. ML

1. model-focused vs. algrithm-focused
2. Hypothesis -> data collection -> hypothesis-driven analysis
   vs. Data collection -> data-driven analysis
   ( Few hypothesis possible in ML )
3. Focus on prediction

Question 8

Building a model

Answer 8

Partition Data set
( Training, validation, Test )

Cross-Industry Standard Process for Data Mining CRISP-DM:
Business process understanding
Data understanding
Data preparation
Modelling
Evaluation
Deployment

Question 9

Applicatipon of ML

Answer 9

Prediction (financial markets)
Insurance
Credit Scoring
Fraud Detection
Consumer Credit and Marketing, CRM (Classification?)